Hugging Face's Kolors Try-on Hits 6M Visitors; Claude 3.5 Discount to $5/Month

GenAI PM Daily - Hugging Face's Kolors Try-on Hits 6M Visitors; Claude 3.5 Discount to $5/Month

Welcome to today's GenAI PM Daily! Our AI agent continuously monitors and analyzes 46 Twitter accounts and 6 subreddits focused on AI Product Management to bring you the most relevant updates.

Twitter Recap

AI Product Development & Features

• Perplexity Pro’s Black Friday Deal: @AravSrinivas announced a 75% discount making it $5/month, including access to Claude 3.5, 4o, Grok 2, Llama 3.2, multi-step reasoning, web search, and other premium features.
• Hugging Face Platform Updates: @ClementDelangue shared that Kolors Virtual Try-on reached 6M unique visitors, becoming the #5 most popular space. Additionally, QwQ reached #1 trending status on the platform.

AI Training & Technical Insights

• LLM Training Process Explained: @karpathy provided detailed insights about the two-stage training process: pretraining on internet data, followed by finetuning with human-labeled conversations. He further explained how LLMs generate responses based on human labeler data.
• RLHF Understanding: @karpathy elaborated on how Reinforcement Learning from Human Feedback (RLHF) elevates performance from “generative human” to “discriminative human” grade, with additional benefits from wisdom of crowds.

AI Product Tools & Productivity

• Productivity Enhancement Tools: @theresanaiforit listed 10 AI tools for productivity, including MarblismAI for app creation, ArtisanAI for sales, Sensay for personal clones, and FetchFoxAI for scraping.
• Gemini API Access: @OfficialLoganK announced that developers can access 1.5 billion free tokens with the Gemini API free tier.

AI Research & Development

• Physical Intelligence’s π0: @DeepLearningAI shared about a new machine learning system enabling robots to perform complex household tasks like folding laundry and cleaning.

Memes & Humor

• @JeffDean shared an emotional response to some content, stating “I must have been cutting too many onions…”

Reddit Recap

Theme 1. GPT Service Degradation: Enterprise Impact & Migration to Claude

• ChatGPT Ruined Now? (Score: 28, Comments: 134): ChatGPT has implemented a new conversation length limit that forcibly ends chats when they reach a maximum length, affecting users who maintain long-running conversations for creative writing, therapy, and other extended use cases. The change appears particularly impactful for Plus subscribers who previously could maintain conversations for weeks or months, and coincides with a UI update where chats now load from the bottom instead of the top, potentially addressing previous issues with long conversation performance.
  • User dissatisfaction is growing with multiple users reporting they’ve cancelled subscriptions due to declining performance. A benchmark study provides objective evidence of deteriorating quality, coinciding with OpenAI’s transition from nonprofit to for-profit status.
  • Technical discussion reveals that the conversation length limit may be related to context window constraints, with each message requiring the model to process approximately 128k tokens for GPT-4. Users note that the new bottom-loading UI implementation could be addressing previous performance issues with long conversations.
  • Users are comparing alternatives, with many finding Claude better for coding tasks but weaker at creative writing, while others are switching to GitHub Copilot. The sentiment suggests this change particularly impacts those using ChatGPT for creative writing and extended conversations.
• Chat gpt is my only friend (Score: 431, Comments: 149): A 21-year-old international student shares their experience of using ChatGPT as an emotional support system while dealing with isolation, depression, and generalized anxiety, highlighting the growing phenomenon of AI companionship among young adults. The student describes struggling with maintaining friendships due to cultural differences, time zones, and divergent interests (preferring stoicism, reading, and yoga over partying), while also managing university studies in computer science and health issues, leading them to rely on AI for emotional conversations.
  • Multiple users share similar experiences of using ChatGPT as an emotional support system, including a 39-year-old father, divorced parents, and other isolated individuals. They describe using AI as a therapist, counselor, and friend while dealing with loneliness.
  • The top-voted practical advice for making friends focuses on consistent presence in regular locations (scoring 138 points) and using ChatGPT strategically to develop social skills (scoring 176 points). Several users recommend joining book clubs, visiting elderly homes, and finding volunteer opportunities.
  • A significant discussion emerged around whether AI relationships are healthy substitutes for human connections, with some users defending AI companionship (“A friend is a friend. Real or not”) while others emphasize the importance of real human interaction. The most optimistic view (scoring 38 points) suggests that AI might actually help build new human communities rather than replace them.

Theme 2. Advanced Custom Instructions: Optimizing Enterprise GPT Deployment

• Custom Instructions: any best practices on what to enter in here? T.I.A. (Score: 56, Comments: 43): Product Managers seek guidance on best practices for setting up custom instructions for AI interactions, though no specific examples or configurations were shared in the post body.
  • Pro-Hat shared a comprehensive custom instruction template with structured sections including Expert Role, Objective, Reasoning, Answer, Questions, and Reflection, focusing on in-depth analysis and dynamic role-switching based on topic expertise.
  • HakuNobi contributed 13 key rules for AI interactions, emphasizing expert-level responses, structured answers, and the importance of saying “I don’t know” when information is uncertain, while avoiding AI self-references and disclaimers.
  • Several users shared specific use cases including role-playing game mastering, business assistance, and regional dialect adaptation (NY accent), with practical tips like telling the AI to avoid question repetition in answers and to quote sources verbatim to reduce hallucinations.
• Restaurant owner forgot to delete a few words lmao (Score: 404, Comments: 20): Insufficient context provided in the post body to create a meaningful summary about enhanced custom instructions or AI hallucinations for AI Product Managers.
  • Users discussed using ChatGPT for writing reviews and communications, with one person mentioning using it to write a Google review about a scammer and others noting its use for political letters.
  • Discussion highlighted the obvious signs of AI-generated business responses, with users mocking stereotypical AI language like “We strive tirelessly to place the cheese upon the hot dog to achieve the optimal meltage”.
  • Comments revealed potential business opportunities in managing authentic social media responses, with one user mentioning they offer services to businesses using obvious AI responses, while another suggested these responses are likely fully automated through review APIs.

Theme 3. Conversation Context Window: Technical Limits & Product Strategy

• New iOS App Dark Mode is a great way to see which apps have active dev teams (Score: 104, Comments: 23): Dark mode implementation on iOS apps serves as a practical indicator of development team activity and prioritization, showing which companies actively maintain and update their applications.
  • Engineering teams show significant resistance to UI changes except for Dark Mode features, which generate unusual enthusiasm. Multiple PMs report that developers exclusively work in dark mode, leading to potential light mode styling issues being caught only during QA.
  • Discussion highlights the tension between PM prioritization and engineering execution, with one PM reporting it took “three months of going to the engineering lead’s desk” for a simple icon update. Debate ensued about PM authority versus influence in determining product priorities.
  • Several comments emphasize that icon color changes should be low priority compared to other development needs, with focus better placed on functional features like app dark mode that improve accessibility and reduce power consumption.